A non-invasive human blood glucose detector and detection method

Abstract

The application develops a human blood glucose noninvasive detector and detection method, relates to biomedical signal collection and processing, realizes collection, feature extraction and blood glucose prediction of human photoelectric plethysmogram signal, and is a low-cost, noninvasive, painless and miniaturized blood glucose meter. The instrument system of the application comprises a clamp unit, a data collection unit, a data processing unit and a display unit, the clamp unit is adjusted to collect pulse wave signals under the optimal detection pressure, the data processing unit predicts blood glucose values and transmits the blood glucose values to the display unit for display. The data processing unit proposes a nonlinear operator integral multi-model decision fusion regression algorithm based on feature optimization weight Euclidean distance, determines the feature optimization weight Euclidean distance of each submodel through the optimal feature matrix of each submodel, and further determines the feature optimization contribution weight of each submodel. The prediction results of each submodel are superimposed through fuzzy integral operator of the feature optimization contribution weight, and the final blood glucose prediction value is obtained.

Description

Technical Field

[0001] This invention relates to the field of biomedical signal acquisition and processing technology, and in particular to a non-invasive human blood glucose detector and detection method. Background Technology

[0002] Diabetes has become the third leading cause of serious health threats among chronic diseases, after cancer and cardiovascular diseases. The clinical, social, and economic impacts of diabetes are primarily related to long-term complications, including cardiovascular events, kidney disease, eye damage, recurrent strokes, and neurological damage. The World Health Organization recommends that diabetic patients monitor their blood glucose levels continuously and in real time, adjusting medical interventions based on the results to effectively control blood glucose levels. Therefore, home blood glucose monitors play a crucial role in both the early diagnosis of diabetes and the management of its complications. Invasive testing methods are convenient and low-cost, and are currently the most widely used home blood glucose testing method. Relying on commercial intermittent blood glucose monitoring devices, they involve collecting blood from the fingertip. The glucose in the blood reacts with the oxidase in the blood glucose test strip, generating a microcurrent that is sent to the microprocessor in the blood glucose meter. The microprocessor then calculates the blood glucose value using mathematical algorithms. This method is traditional and currently the most common blood glucose testing method on the market, providing relatively accurate results. However, it requires blood collection, creating a wound that is prone to infection. Furthermore, the fingertip is the most common source of blood, and the dense distribution of nerve endings in the fingertip can cause pain and psychological aversion for users. Therefore, low-cost, non-invasive home blood glucose testing devices that do not cause any trauma to test subjects have become the focus of researchers' efforts.

[0003] Non-invasive detection methods can be divided into two categories: optical and non-optical. Applying near-infrared spectroscopy for non-invasive blood glucose detection is a focus of many research teams. However, near-infrared non-invasive blood glucose detection technology faces challenges such as weak signals and strong background interference, which need to be addressed. Furthermore, near-infrared spectrometers are expensive and bulky, making them unsuitable for portable, miniaturized, and low-cost applications. In contrast, photoplethysmography (PPG) offers advantages such as simple signal acquisition, low cost, and high correlation with blood glucose concentration, making it highly suitable for developing low-cost non-invasive blood glucose detection devices.

[0004] Based on this, the present invention develops a non-invasive blood glucose meter and method based on photoplethysmography (PPG) signals, which realizes the acquisition, feature extraction and blood glucose value prediction of human PPG signals. It is a low-cost, non-invasive, painless, miniaturized and high-precision blood glucose meter. Summary of the Invention

[0005] The purpose of this invention is to provide a non-invasive human blood glucose detector and detection method. It collects pulse wave signals reflected from the test site by an LED light source, extracts signal features, and establishes a quantitative prediction model to non-invasively detect human blood glucose concentration. It proposes a new nonlinear operator integral multi-model decision fusion regression algorithm based on feature optimization weight Euclidean distance. This method determines the feature optimization weight Euclidean distance of each sub-model through features, and then determines the feature optimization contribution weight of each sub-model, thereby improving the accuracy of blood glucose concentration prediction.

[0006] To achieve the above objectives, the present invention provides the following solution:

[0007] A non-invasive blood glucose monitoring device for humans, comprising:

[0008] The clamping unit is used to clamp and fix the part of the human body to be measured, and the optimal measurement pressure value is achieved by adjusting the clamping unit.

[0009] The data acquisition unit is used to acquire the photoplethysmography (PPG) signal of the measured part.

[0010] The data processing and control unit, connected to the data acquisition unit, is used to receive the photoplethysmography (PPG) signal from the data acquisition unit, and to perform feature extraction and multivariate quantitative regression prediction on the PPG signal to obtain the blood glucose value of the subject.

[0011] Optionally, the detector further includes an input / output unit connected to the data processing and control unit for displaying the subject's blood glucose level.

[0012] Optionally, the clamping unit specifically includes:

[0013] Pressure sensor, temperature sensor, first A / D conversion module, finger fixing slot, LED light source hole, detector hole, threaded hole, acquisition end circuit board placement area, spring hole, Velcro fixing bracket, spring, pressure sensor placement area, spring fixing hole and housing limit bracket;

[0014] The pressure sensor is disposed in the pressure sensor placement area and is connected to the data processing and control unit for detecting the pressure of the measured part.

[0015] The temperature sensor is connected to the data processing and control unit and is used to detect the body temperature of the subject.

[0016] The first A / D conversion module is connected to the pressure sensor, temperature sensor and data processing and control unit, and is used to convert the detected pressure and temperature into electrical signals and transmit them to the data processing and control unit.

[0017] The finger fixing groove is used to fix the position of the subject's finger so that the finger is in contact with the pressure sensor;

[0018] Threaded holes are provided around the finger fixing groove and the area for placing the acquisition terminal circuit board; the threaded holes are used to connect the finger fixing groove and the area for placing the acquisition terminal circuit board.

[0019] The finger fixing groove is also provided with an LED light source hole and a detector hole;

[0020] A spring hole is provided below the placement area of ​​the acquisition terminal circuit board, and a spring is provided in the spring hole. The spring and the pressure sensor are combined to maintain the consistency of the measurement position and pressure, and ensure that the optimal detection pressure is achieved.

[0021] A Velcro fixing bracket is also provided below the area where the acquisition terminal circuit board is placed;

[0022] The data processing and control unit is used to determine whether the received pressure electrical signal meets the requirements. If it does, it starts to collect the photoplethysmography (PPG) signal of the measured part. If it does not meet the requirements, it adjusts the pressure until the requirements are met.

[0023] The data processing and control unit is also used to send the pressure value to the input / output unit for display.

[0024] Optionally, the data acquisition unit specifically includes:

[0025] Constant current source drive circuit, LED light source group and photodetector group;

[0026] The constant current source drive circuit is used to start the LED light sources in the LED light source group in turn and at different times.

[0027] The LED light source group is used to emit near-infrared light to the part of the human body being measured;

[0028] The photodetector array is used to receive the light signal reflected from the part of the human body being measured, thereby generating a microcurrent signal, and sending the microcurrent signal to the data processing unit.

[0029] Optionally, the LED light source group includes: two measuring light sources and two reference light sources;

[0030] The photodetector group includes: a Si photodetector and an InGaAs photodetector.

[0031] Optionally, the data processing unit includes:

[0032] Signal conditioning circuit, second A / D conversion module, and microcontroller (MCU);

[0033] The signal conditioning circuit is used to perform IV conversion, amplification, and filtering on the signals acquired by the data acquisition unit.

[0034] The second A / D conversion module is used to convert the signal after IV conversion, amplification and filtering into an electrical signal and send it to the microcontroller MCU;

[0035] The microcontroller (MCU) is used to extract features from the received signal and perform multivariate quantitative regression prediction to obtain the blood glucose value of the subject.

[0036] Optionally, the signal conditioning circuit includes: an IV conversion circuit, an adaptive amplifier circuit, a high-pass filter, and a low-pass filter connected in sequence.

[0037] Optionally, the step of performing feature extraction and multivariate quantitative regression prediction on the photoplethysmography signal to obtain the subject's blood glucose value specifically includes:

[0038] Feature extraction is performed on the photoplethysmography (PPG) signal to obtain the feature vector: [R] 1 / 3 ,R 1 / 4 ,R 2 / 3 ,R 2 / 4 [SE1,C01,FD1,PE1,SE2,C02,FD2,PE2,SE3,C03,FD3,PE3,SE4,C04,FD4,PE4]; where the subscripts 1, 2, 3, and 4 represent the characteristics of the PPG signal collected by LEDs of different wavelengths.

[0039] Obtain the training set N X×Y And M sub-models; the training set includes X sets of training data, each set containing Y features, and the training set ground truth labels are L. X ;

[0040] After conducting guided trial-and-error experiments with M sub-models, different feature values ​​are randomly selected for each combination prediction. The feature combination that minimizes the root mean square error between the prediction results of the training set and the true value is determined, and the feature matrix N most suitable for the M seed model algorithm is obtained. X×Y1 ', N X×Y2 ',N X×Y3 ',…,N X×YM ' and M-seed models respectively predict the training set value P under the optimal features. X1 ', P X2 ', P X3 ',…,P XM The predicted values ​​are then arranged into a non-decreasing sequence, and the rearranged predicted value is P. X1 ,P X2 ,P X3 ,…,P XM The corresponding feature matrix NX×Y1 N X×Y2 N X×Y3 ,…,N X×YM ;

[0041] Calculate the truth value L X The prediction results P of each sub-model X1 P X2 P X3 Feature optimization weights and Euclidean distance d(i):

[0042]

[0043]

[0044] Where, β i R represents the fuzzy adjustment parameter. i T represents the number of times the feature value contained in the i-th sub-model appears in all sub-models. a Let P be the number of times the a-th feature of the optimal feature matrix of the i-th sub-model is used in all sub-models. Xi Yi represents the prediction result of the Xth training sample in the i-th sub-model; Yi represents the number of eigenvalues ​​in the optimal feature matrix of the i-th sub-model.

[0045] The feature optimization contribution weights of each sub-model are calculated based on the Euclidean distance of the feature optimization weights, and normalization is used to ensure that the sum of the weights is 1. The calculation formula is as follows:

[0046]

[0047]

[0048] Feature extraction is performed on the acquired PPG signal to obtain the feature vector: [R] 1 / 3 ,R 1 / 4 ,R 2 / 3 ,R 2 / 4 [SE1,C01,FD1,PE1,SE2,C02,FD2,PE2,SE3,C03,FD3,PE3,SE4,C04,FD4,PE4]; where the subscripts 1, 2, 3, and 4 represent the characteristics of PPG signals acquired by LEDs with wavelengths of 878-882nm, 940-945nm, 1200-1205nm, and 1550-1554nm, respectively;

[0049] The feature vectors are substituted into each sub-model, and then the nonlinear operator is introduced through the trained weights to obtain the blood glucose prediction value.

[0050] Optionally, the method of integrating the trained weights with a nonlinear operator to obtain the blood glucose prediction value specifically uses the following formula:

[0051]

[0052] Where C(P) is the predicted blood glucose value, M is the number of sub-models, and ω t Indicates the weight.

[0053] Secondly, the present invention provides a non-invasive method for detecting human blood glucose, the method comprising:

[0054] Turn on the power to supply power to all units of the system and its own functional modules;

[0055] Perform system initialization operations to bring the system into working state and assign initial state values ​​to each module of the system;

[0056] In the initial state, the part to be measured is fixed by an adjustable clamp;

[0057] The four LED light sources are started in turn by a constant current source driving circuit, with each light source lasting for 3 seconds and each detection time lasting 12 seconds. The wavelength ranges of the LED light sources are 878-882nm, 940-945nm, 1200-1205nm and 1550-1554nm, respectively.

[0058] A high-wavelength LED light source emits stable near-infrared light, which, after being reflected by the part being measured, reaches an InGaAs photodetector that can respond to the near-infrared band; the high-wavelength band is 1200-1205nm and 1550-1554nm.

[0059] The light emitted by the low-wavelength LED light source is reflected by the part being measured and then reaches the Si photodetector; the low-wavelength range is 878-882nm and 940-945nm.

[0060] After receiving the light signal reflected from the measured part, the InGaAs photodetector and the Si photodetector generate a micro-current signal.

[0061] The microcurrent signal is first converted to IV, then amplified and filtered, and then converted to AD before being input to the microcontroller MCU. The microcontroller processes the acquired signal to obtain the subject's blood glucose value.

[0062] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0063] This invention proposes a nonlinear operator integral multi-model decision fusion regression algorithm based on feature optimization weight Euclidean distance. It establishes a formula for calculating feature optimization weights, determining the feature optimization weights of each sub-model by calculating the Euclidean distance of the feature optimization weights of each sub-model. This enhances the contribution of the optimal features to the final prediction model, improving the overall prediction accuracy and generalization ability of the algorithm. Furthermore, this invention proposes a novel fuzzy operator integral strategy. The prediction results of each sub-model are rearranged and mapped to fuzzy operators. Based on the rearranged prediction results and the corresponding weights of each sub-model, the fuzzy operators are nonlinearly superimposed, reducing the mutual influence of overlapping features used by each sub-model and strengthening the contribution of high-precision sub-models, thus improving the prediction accuracy of the algorithm. This invention proposes four LEDs with specific wavelengths as light sources, rich in blood glucose concentration information. Using LEDs as light sources allows for time-division multiplexing, resulting in low hardware cost, small size, and convenient and fast operation, meeting the requirements of low cost, miniaturization, and portability. In addition, compared with commonly used invasive and minimally invasive blood glucose testing devices, this invention adopts a non-invasive testing method, which does not cause pain to the user and has no risk of infection; the clamp unit uses a combination of spring structure and pressure sensor to maintain the consistency of measurement position and pressure, and ensure that the optimal detection pressure is achieved, eliminating pressure interference and improving prediction accuracy. Attached Figure Description

[0064] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0065] Figure 1 This is a schematic diagram illustrating the principle of the non-invasive human blood glucose detector provided by the present invention.

[0066] Figure 2 This is a schematic diagram of the fixture unit structure provided by the present invention;

[0067] Figure 3 A flowchart illustrating the blood glucose measurement process of the non-invasive human blood glucose detector provided by the present invention.

[0068] Figure 4 A schematic diagram of the pulse wave signal acquired by this invention;

[0069] Figure 5 A schematic diagram of blood glucose concentration in a subject provided by the present invention;

[0070] Figure 6 The subjects provided by this invention use the Clarke error grid results of the predicted blood glucose value and the true blood glucose value of this invention. Detailed Implementation

[0071] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0072] The purpose of this invention is to provide a non-invasive human blood glucose detector and detection method. It collects pulse wave signals reflected from the test site by an LED light source, extracts signal features, and establishes a quantitative prediction model to non-invasively detect human blood glucose concentration. It proposes a new nonlinear operator integral multi-model decision fusion regression algorithm based on feature optimization weight Euclidean distance. This method determines the feature optimization weight Euclidean distance of each sub-model through features, and then determines the feature optimization contribution weight of each sub-model, thereby improving the accuracy of blood glucose concentration prediction.

[0073] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0074] Example 1

[0075] Figure 1 This is a schematic diagram illustrating the principle of the non-invasive human blood glucose meter provided by the present invention, as shown below. Figure 1 As shown, the non-invasive blood glucose monitoring device of the present invention includes:

[0076] Fixture unit, data acquisition unit, data processing unit, and display unit;

[0077] See Figure 2 The fixture unit includes: a pressure sensor, a temperature sensor, a first A / D conversion module, a finger fixing slot 2, an LED light source hole, a detector hole, a threaded hole 4, a data acquisition end circuit board placement area 3, a spring hole 5, a Velcro fixing bracket 6, a spring, a pressure sensor placement area 7, a spring fixing hole 8, and a housing limiting bracket 9. Figure 2 The number 1 in the diagram represents the LED light source hole and the detector hole;

[0078] The pressure sensor is located in the pressure sensor placement area 7 and is connected to the data processing and control unit for detecting the pressure of the measured part.

[0079] Specifically, the finger fixing groove is used to clamp and fix the test part of the human body to reproduce the measurement pressure to the ideal value, so as to achieve the best measurement pressure effect. The optimal pressure effect is within the range of 3.9-5N. Exceeding this pressure range will lead to larger measurement errors. Specifically, the pressure sensor collects the pressure of the test part, and the measured pressure is converted into an electrical signal by the first A / D conversion module and transmitted to the microcontroller (MCU). The microcontroller determines whether the requirements are met. If they are met, the signal is transmitted to the display unit for display, and the next data acquisition operation is performed. If the requirements are not met, the display unit displays corresponding prompts to prompt the subject to adjust the pressure until the pressure requirement is met.

[0080] The temperature sensor is connected to the data processing and control unit and is used to detect the body temperature of the subject.

[0081] The first A / D conversion module is connected to the pressure sensor, temperature sensor and data processing and control unit, and is used to convert the detected pressure and temperature into electrical signals and transmit them to the data processing and control unit.

[0082] The finger fixing groove 2 is used to fix the position of the subject's finger so that the finger is in contact with the pressure sensor;

[0083] Threaded holes 4 are provided around the finger fixing groove 2 and the acquisition end circuit board placement area 3; the threaded holes 4 are used to connect the finger fixing groove 2 and the acquisition end circuit board placement area 3.

[0084] The finger fixing groove is also provided with an LED light source hole and a detector hole;

[0085] A spring hole 5 is provided below the circuit board placement area 3 of the acquisition end, and a spring is provided in the spring hole 5. The spring and the pressure sensor are combined to maintain the consistency of the measurement position and pressure, and ensure that the optimal detection pressure is achieved.

[0086] Below the circuit board placement area 3 of the acquisition terminal is a Velcro fixing bracket 6;

[0087] The data processing and control unit is used to determine whether the received pressure electrical signal meets the requirements. If it does, it starts to collect the photoplethysmography (PPG) signal of the measured part. If it does not meet the requirements, it adjusts the pressure until the requirements are met.

[0088] The data processing and control unit is also used to send the pressure value to the input / output unit for display.

[0089] The data acquisition unit specifically includes: a constant current source drive circuit, an LED light source group, and a photodetector group; wherein, the LED light source group includes: two measurement light sources and two reference light sources; the photodetector group includes: a Si photodetector and an InGaAs photodetector.

[0090] The data processing unit includes: a signal conditioning circuit, a second A / D conversion module, and a microcontroller (MCU); the signal conditioning circuit includes: an IV conversion circuit, an adaptive amplifier circuit, a high-pass filter, and a low-pass filter connected in sequence.

[0091] The LED light sources in the data acquisition unit have wavelengths of 880nm, 940nm, 1200nm, and 1550nm, respectively. The LED light sources are used to provide light source conditions to facilitate data acquisition by the photodetector.

[0092] After receiving the light signal reflected from a finger or other part of the body being measured, the photodetector generates a weak current signal. To facilitate subsequent acquisition and processing, the signal is first converted from an IV signal by an IV converter circuit, then amplified by an adaptive amplifier circuit, and finally filtered by a 0.1Hz high-pass filter and a 10Hz low-pass filter. After being converted into an electrical signal by the second A / D conversion module, the signal is finally transmitted to the microcontroller. The microcontroller performs feature extraction, quantitative regression prediction, and other processing on the signal to obtain the predicted blood glucose value. Finally, the measured blood glucose value is transmitted to the display unit.

[0093] The microcontroller's processing procedure is as follows:

[0094] The acquired PPG signal was subjected to spatiotemporal dual-modal feature extraction to obtain the feature vector: [R 1 / 3 ,R 1 / 4 ,R 2 / 3 ,R 2 / 4 [SE1,C01,FD1,PE1,SE2,C02,FD2,PE2,SE3,C03,FD3,PE3,SE4,C04,FD4,PE4]; where the subscripts 1, 2, 3, and 4 represent the features of PPG signals collected by LED light sources with wavelengths of 880nm, 1550nm, 940nm, and 1200nm, respectively; the feature vectors are passed to the quantitative prediction module, which is a pre-trained model.

[0095] The extracted features include: R coefficient features, spectral entropy (SE), C0 complexity, permutation entropy (PE), and fractal dimension (FD).

[0096] The R coefficient is derived from the Lambert-Beer law:

[0097] λ1 is used as the measurement light, and another light source with wavelength λ2 is used as the reference light. In this invention, the measurement light is 880nm and 1550nm wavelength light, and the reference light is 1200nm and 940nm wavelength light.

[0098] The spectral entropy (SE) is calculated as follows:

[0099] a. Remove DC and calculate the mean of the PPG signal sequence x(n). Subtract the mean from the sequence one by one, and do...

[0100] b. Perform a discrete Fourier transform on the sequence x(n) to obtain K(n).

[0101] c. Calculate the relative power spectrum. According to Paserval's theorem, calculate the power spectrum value at a certain frequency point:

[0102]

[0103] The relative power spectral probability P(n) of the sequence is:

[0104]

[0105] d. Calculate the spectral entropy SE:

[0106]

[0107] The C0 complexity is calculated as follows:

[0108] a. Perform a discrete FFT transform on the sequence:

[0109]

[0110] b. Remove the irregular part of X(k), keep the regular part, and add the parameter r to the mean square of X(k), keep the part that exceeds r times the mean square, and set the value of the rest to 0.

[0111]

[0112] c. Perform an inverse Fourier transform on the above equation.

[0113]

[0114] d. The complexity of C0 is:

[0115]

[0116] The permutation entropy PE is calculated as follows:

[0117] a. Set a time delay parameter τ, downsample the sequence, and sample every τ data points to create a new one-dimensional array: X(i)=[x(i),x(i+τ),...,x(i+τ(m-1))], where τ is 3 in this embodiment.

[0118] b. Sort each element of the sequence X(i) and write it into a new array D(i), which represents the position of the element X(i) in the array X.

[0119] c. For an m-dimensional vector, there are m! possible permutations. Define the probability of each permutation.

[0120]

[0121] d. Calculate the permutation entropy:

[0122]

[0123] The fractal dimension FD is calculated as follows:

[0124] The dimension formula means determining the dimension of the shape by covering it with a small cube of side length ε. The calculation formula is as follows; in this embodiment, ε is chosen to be 3.

[0125]

[0126] The training process of the model is as follows:

[0127] Step 1: The training set has N features X×Y The dataset contains X sets of training data, each set containing Y features. In this embodiment, Y is 20, and the ground truth labels for the training set are L. i Through guided trial and error experiments using multiple sub-models, this invention selects random forest, extreme learning machine, and gradient boosting algorithms, which have high prediction accuracy. Different feature values ​​are randomly selected for combination prediction each time, until all feature combinations are traversed. The feature combination that minimizes the root mean square standard deviation of the predicted results on the training set is determined.

[0128]

[0129] In the formula, L i ' is the predicted value, L i This is the true blood glucose value.

[0130] The feature matrix N best suited for random forest, extreme learning machine, and gradient boosting model algorithms are obtained respectively. X×Y1 ', N X×Y2 ',N X×Y3', and the training set prediction values ​​P of the three sub-models under the optimal features respectively. X1 ', P X2 ', P X3 The predicted values ​​are then arranged into a non-decreasing sequence, and the rearranged predicted value is P. X1 P X2 P X3 The corresponding feature matrix N X×Y1 N X×Y2 N X×Y3 .

[0131] Step 2: Based on the feature optimization weight method, the prediction results of each model are fused using nonlinear operator integration. First, the concept of feature optimization weight is introduced, and the true value L is calculated. X The prediction results P of each sub-model X1 P X2 P X3 Feature optimization weights Euclidean distance d(i), where β i It belongs to the fuzzy adjustment parameter, which is determined based on the number of features contained in each sub-model and the number of times the contained features are used in all sub-models.

[0132]

[0133]

[0134] In the formula R i T represents the number of times the feature value contained in the i-th sub-model appears in all sub-models. a The a-th feature of the optimal feature matrix of the i-th sub-model is used the most times in all sub-models.

[0135] Step 3: Calculate the feature optimization contribution weights of each sub-model based on the Euclidean distance of the feature optimization weights, and ensure that the sum of the weights is 1 by normalization.

[0136]

[0137]

[0138] Step 4: Extract features from the acquired PPG signals, substitute the features into each sub-model, and then integrate them using trained weights and a nonlinear operator. The integral can be expressed as:

[0139]

[0140] Where C(P) is the predicted value, M is the number of sub-models (3 in this embodiment), and ω is the weight. According to the requirements of the nonlinear operator integral operator, let P... X0 =0.

[0141] Example 2

[0142] Figure 3 This is a schematic diagram illustrating the process of measuring human blood glucose using the non-invasive human blood glucose meter of the present invention, as shown below. Figure 3 As shown, the process includes:

[0143] Step 1: Turn on the power to supply power to all units of the system and its own functional modules.

[0144] Step 2: Perform system initialization operations to put the system into working state and assign initial state values ​​to each module of the system.

[0145] Step 3: In the initial state, fix the part to be measured with the adjustable clamp to reproduce the measurement pressure to the ideal value.

[0146] Here, the ideal value is obtained through prior simulation calculations and experimental verification to determine the optimal detection pressure. First, the subject's finger is fixed in position using a finger fixation groove. Then, the measurement pressure is increased or decreased using a spring, and the display screen is observed until the measurement requirements are met, driving the acquisition of the PPG signal.

[0147] Step 3: PPG Signal Acquisition: Four light sources are activated in turn using a constant current source drive circuit, with each light source activating for 3 seconds. Therefore, each detection time is 12 seconds. The 1550nm and 1200nm LED light sources emit stable near-infrared light, which, after reflection from the fingertip, reaches the InGaAs photodetector that can respond to the near-infrared band. The 880nm and... 940nm The light emitted by the LED light source is reflected by the fingertip and reaches the Si-based photodetector. The photodetector receives the reflected light signal and generates a weak current signal. For easier subsequent acquisition and processing, an IV-to-IV conversion is performed, followed by amplification. The acquired pulse wave signal is illustrated in the diagram below. Figure 4 And input signal conditioning circuit, input to MCU minimum system.

[0148] Step 4: Extract features from the above signals and substitute them into the established model to calculate the blood glucose level of the subject.

[0149] To evaluate the accuracy and error of the non-invasive blood glucose monitoring system, it is necessary to use a stable invasive blood glucose meter on the market to obtain the blood glucose concentration of the human body as a reference value. This invention uses the OneSmart portable blood glucose meter manufactured by Johnson & Johnson Medical to obtain the true blood glucose values ​​of the subjects. A two-day calibration experiment was conducted, collecting 12 samples. The subjects were all healthy adult males who fasted for 12 hours before the experiment. The equipment was debugged, and communication with the host computer was established via serial port. Samples were collected in the morning, noon, and evening, before and after meals. The collected data are as follows: Figure 5 As shown.

[0150] Before measurement, the pressure sensor in the clamp ensures the subject reaches optimal detection pressure before acquiring the PPG signal. During the acquisition process, the subject should remain emotionally stable, avoid strenuous exercise, and fully utilize the system's structural components to ensure close contact between the finger and the photoelectric device, guaranteeing the accuracy and repeatability of each sample. After acquiring the PPG spectral data using the non-invasive blood glucose detection system, the true blood glucose concentration at the corresponding moment should be obtained as soon as possible using an invasive blood glucose meter to avoid the influence of time deviation.

[0151] After acquiring PPG signals, spatiotemporal dual-modal feature extraction was performed, and then blood glucose concentration was predicted through a quantitative regression model. In the prediction module, a fuzzy integral quantitative multi-model decision fusion method was proposed. This method determines the feature optimization weights of each sub-model through the Euclidean distance of the features, and then determines the feature optimization contribution weights of each sub-model, thereby improving the accuracy of blood glucose concentration prediction.

[0152] This invention introduces Clarke Error Grid Analysis (EGA) to further evaluate the system's ability to detect blood glucose concentration based on the predicted root mean square error. The system test results are shown in the table below:

[0153]

[0154] The RMSE of the test result was 1.14 mmol / L, which was determined by... Figure 6 As can be seen, except for sample 4 on the first day which showed a significant deviation, there was a good correlation between the reference and predicted blood glucose values. The table shows that, except for samples 1, 3, and 4 on the first day, the absolute values ​​of the relative errors for the remaining samples were all within 20%. Clarke error grid analysis indicates that 75% of the samples were in region A, and no samples entered regions C, D, or E. In summary, the system has good performance in blood glucose prediction, and its accuracy basically meets the requirements of practical measurement.

[0155] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0156] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A non-invasive human blood glucose testing device, characterized in that, include: The clamping unit is used to clamp and fix the part of the human body to be measured, and the clamping unit is adjusted to achieve the optimal measurement pressure value; The data acquisition unit is used to acquire the photoplethysmography (PPG) signal of the measured part. A data processing and control unit, connected to the data acquisition unit, is used to receive photoplethysmography (PPG) signals from the data acquisition unit, and to perform feature extraction and multivariate quantitative regression prediction on the PPG signals to obtain the blood glucose value of the subject. Feature extraction and multivariate quantitative regression prediction of the photoplethysmography (PPG) signal are performed to obtain the subject's blood glucose value, specifically including: Feature extraction is performed on the photoelectric plethysmogram signal to obtain a feature vector: [R 1 / 3 , 1 / 4 , 2 / 3 , 2 / 4 , SE1, C01, FD1, PE1, SE2, C02, FD2, PE2, SE3, C03, FD3, PE3, SE4, C04, FD4, PE4]; wherein subscripts 1, 2, 3, and 4 respectively represent features of PPG signals collected by LEDs of different wavelengths; R represents a coefficient feature, SE represents spectral entropy, C0 represents complexity, PE represents permutation entropy, and FD represents a fractal dimension. Obtain the training set and M sub-models; the training set includes X sets of training data, each set containing Y features, and the training set ground truth labels are... ; After conducting guided trial-and-error experiments with M sub-models, different feature values ​​are randomly selected for each combination prediction. The feature combination that minimizes the root mean square error between the prediction results of the training set and the true value is determined, and the feature matrix most suitable for the M sub-model algorithms is obtained. , , ,…, , and the training set predictions of the M sub-models under the optimal features , , ,…, The predicted values ​​are then arranged into a non-decreasing sequence, and the rearranged predicted values ​​are... , , ,…, Corresponding feature matrix , , ,…, ; Calculate the truth value Prediction results of each sub-model , , Feature optimization weight Euclidean distance d ( i ): ； ； in, β i Indicates fuzzy adjustment parameters. R i For the first i The number of times each feature value contained in a sub-model appears in all sub-models. T a For the first i The optimal feature matrix of the sub-model a The number of times a feature is used in all sub-models P Xi Indicates the first X The training sample at the th ... i Prediction results in each sub-model; Yi Indicates the first i The number of eigenvalues ​​in the optimal feature matrix of each sub-model; The feature optimization contribution weights of each sub-model are calculated based on the Euclidean distance of the feature optimization weights, and normalization is used to ensure that the sum of the weights is 1. The calculation formula is as follows: ； in, This represents the feature optimization contribution weight of each sub-model; This represents the feature optimization contribution weights before normalization. Feature extraction is performed on the acquired PPG signal to obtain the feature vector: [R] 1 / 3 ,R 1 / 4 ,R 2 / 3 ,R 2 / 4 [SE1,C01,FD1,PE1,SE2,C02,FD2,PE2,SE3,C03,FD3,PE3,SE4,C04,FD4,PE4]; where the subscripts 1, 2, 3, and 4 represent the characteristics of PPG signals acquired by LEDs with wavelengths of 878-882nm, 940-945nm, 1200-1205nm, and 1550-1554nm, respectively; The feature vectors are substituted into each sub-model, and then the nonlinear operator is introduced through the trained weights to obtain the blood glucose prediction value.

2. The non-invasive human blood glucose meter according to claim 1, characterized in that, The detector also includes an input / output unit connected to the data processing and control unit for displaying the subject's blood glucose level.

3. The non-invasive human blood glucose meter according to claim 1, characterized in that, The fixture unit specifically includes: Pressure sensor, temperature sensor, first A / D conversion module, finger fixing slot, LED light source hole, detector hole, threaded hole, acquisition end circuit board placement area, spring hole, Velcro fixing bracket, spring, pressure sensor placement area, spring fixing hole and housing limit bracket; The pressure sensor is disposed in the pressure sensor placement area and is connected to the data processing and control unit for detecting the pressure of the measured part. The temperature sensor is connected to the data processing and control unit and is used to detect the body temperature of the subject. The first A / D conversion module is connected to the pressure sensor, temperature sensor and data processing and control unit, and is used to convert the detected pressure and temperature into electrical signals and transmit them to the data processing and control unit. The finger fixing groove is used to fix the position of the subject's finger so that the finger is in contact with the pressure sensor; Threaded holes are provided around the finger fixing groove and the area for placing the acquisition terminal circuit board; the threaded holes are used to connect the finger fixing groove and the area for placing the acquisition terminal circuit board. The finger fixing groove is also provided with an LED light source hole and a detector hole; A spring hole is provided below the placement area of ​​the acquisition terminal circuit board, and a spring is provided in the spring hole. The spring and the pressure sensor are combined to maintain the consistency of the measurement position and pressure, and ensure that the optimal detection pressure is achieved. A Velcro fixing bracket is also provided below the area where the acquisition terminal circuit board is placed; The data processing and control unit is used to determine whether the received pressure electrical signal meets the requirements. If it does, it starts to collect the photoplethysmography (PPG) signal of the measured part. If it does not meet the requirements, it adjusts the pressure until the requirements are met. The data processing and control unit is also used to send the pressure value to the input / output unit for display.

4. The non-invasive human blood glucose meter according to claim 1, characterized in that, The data acquisition unit specifically includes: Constant current source drive circuit, LED light source group and photodetector group; The constant current source drive circuit is used to start the LED light sources in the LED light source group in turn and at different times. The LED light source group is used to emit near-infrared light to the part of the human body being measured; The photodetector array is used to receive the light signal reflected from the part of the human body being measured, thereby generating a microcurrent signal, and sending the microcurrent signal to the data processing unit.

5. The non-invasive human blood glucose detector according to claim 4, characterized in that, The LED light source group includes: two measuring light sources and two reference light sources; The photodetector group includes: a Si photodetector and an InGaAs photodetector.

6. The non-invasive human blood glucose detector according to claim 1, characterized in that, The data processing unit includes: Signal conditioning circuit, second A / D conversion module, and microcontroller (MCU); The signal conditioning circuit is used to perform IV conversion, amplification, and filtering on the signals acquired by the data acquisition unit. The second A / D conversion module is used to convert the signal after IV conversion, amplification and filtering into an electrical signal and send it to the microcontroller MCU; The microcontroller (MCU) is used to extract features from the received signal and perform multivariate quantitative regression prediction to obtain the blood glucose value of the subject.

7. The non-invasive human blood glucose meter according to claim 6, characterized in that, The signal conditioning circuit includes: an IV conversion circuit, an adaptive amplifier circuit, a high-pass filter, and a low-pass filter connected in sequence.

8. The non-invasive human blood glucose detector according to claim 1, characterized in that, The blood glucose prediction value is obtained by integrating a nonlinear operator using trained weights, specifically using the following formula: ； in, It is a blood glucose prediction value. M This is the number of sub-models, M=3. This represents the feature optimization contribution weight of each sub-model.

9. A non-invasive method for detecting human blood glucose, characterized in that, The detection method includes: Turn on the power to supply power to all units of the system and its own functional modules; Perform system initialization operations to bring the system into working state and assign initial state values ​​to each module of the system; In the initial state, the part to be measured is fixed by an adjustable clamp; The four LED light sources are started in turn by a constant current source driving circuit, with each light source lasting for 3 seconds and each detection time lasting 12 seconds. The wavelength ranges of the LED light sources are 878-882nm, 940-945nm, 1200-1205nm and 1550-1554nm, respectively. A high-wavelength LED light source emits stable near-infrared light, which, after being reflected by the part being measured, reaches an InGaAs photodetector that can respond to the near-infrared band; the high-wavelength band is 1200-1205nm and 1550-1554nm. The light emitted by the low-wavelength LED light source is reflected by the part being measured and then reaches the Si photodetector; the low-wavelength range is 878-882nm and 940-945nm. After receiving the light signal reflected from the measured part, the InGaAs photodetector and the Si photodetector generate a micro-current signal. The microcurrent signal is first converted to IV, then amplified and filtered, and then converted to AD before being input to the microcontroller MCU. The microcontroller processes the acquired signal to obtain the blood glucose value of the subject. Feature extraction and multivariate quantitative regression prediction of the photoplethysmography (PPG) signal are performed to obtain the subject's blood glucose value, specifically including: Feature extraction is performed on the photoplethysmography (PPG) signal to obtain the feature vector: [R] 1 / 3 ,R 1 / 4 ,R 2 / 3 ,R 2 / 4 [SE1,C01,FD1,PE1,SE2,C02,FD2,PE2,SE3,C03,FD3,PE3,SE4,C04,FD4,PE4]; where subscripts 1, 2, 3, and 4 represent the characteristics of PPG signals acquired by LEDs at different wavelengths; R represents coefficient characteristics, SE represents spectral entropy, C0 represents complexity, PE represents permutation entropy, and FD represents fractal dimension; Obtain the training set and M sub-models; the training set includes X sets of training data, each set containing Y features, and the training set ground truth labels are... ; After conducting guided trial-and-error experiments with M sub-models, different feature values ​​are randomly selected for each combination prediction. The feature combination that minimizes the root mean square error between the prediction results of the training set and the true value is determined, and the feature matrix most suitable for the M sub-model algorithms is obtained. , , ,…, , and the training set predictions of the M sub-models under the optimal features , , ,…, The predicted values ​​are then arranged into a non-decreasing sequence, and the rearranged predicted values ​​are... , , ,…, Corresponding feature matrix , , ,…, ; Calculate the truth value Prediction results of each sub-model , , Feature optimization weight Euclidean distance d ( i ): ； ； in, β i Indicates fuzzy adjustment parameters. R i For the first i The number of times each feature value contained in a sub-model appears in all sub-models. T a For the first i The optimal feature matrix of the sub-model a The number of times a feature is used in all sub-models P Xi Indicates the first X The training sample at the th ... i Prediction results in each sub-model; Yi This represents the number of eigenvalues ​​in the optimal feature matrix of the i-th sub-model; The feature optimization contribution weights of each sub-model are calculated based on the Euclidean distance of the feature optimization weights, and normalization is used to ensure that the sum of the weights is 1. The calculation formula is as follows: ； in, This represents the feature optimization contribution weight of each sub-model; This represents the feature optimization contribution weights before normalization. Feature extraction is performed on the acquired PPG signal to obtain the feature vector: [R] 1 / 3 ,R 1 / 4 ,R 2 / 3 ,R 2 / 4 [SE1,C01,FD1,PE1,SE2,C02,FD2,PE2,SE3,C03,FD3,PE3,SE4,C04,FD4,PE4]; where the subscripts 1, 2, 3, and 4 represent the characteristics of PPG signals acquired by LEDs with wavelengths of 878-882nm, 940-945nm, 1200-1205nm, and 1550-1554nm, respectively; The feature vectors are substituted into each sub-model, and then the nonlinear operator is introduced through the trained weights to obtain the blood glucose prediction value.

Images